]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Syncookies implementation for the Linux kernel | |
3 | * | |
4 | * Copyright (C) 1997 Andi Kleen | |
e905a9ed | 5 | * Based on ideas by D.J.Bernstein and Eric Schenk. |
1da177e4 LT |
6 | * |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; either version | |
10 | * 2 of the License, or (at your option) any later version. | |
1da177e4 LT |
11 | */ |
12 | ||
13 | #include <linux/tcp.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/random.h> | |
16 | #include <linux/cryptohash.h> | |
17 | #include <linux/kernel.h> | |
18 | #include <net/tcp.h> | |
86b08d86 | 19 | #include <net/route.h> |
1da177e4 | 20 | |
4dfc2817 FW |
21 | /* Timestamps: lowest 9 bits store TCP options */ |
22 | #define TSBITS 9 | |
23 | #define TSMASK (((__u32)1 << TSBITS) - 1) | |
24 | ||
1da177e4 LT |
25 | extern int sysctl_tcp_syncookies; |
26 | ||
2051f11f | 27 | __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; |
c6aefafb | 28 | EXPORT_SYMBOL(syncookie_secret); |
1da177e4 LT |
29 | |
30 | static __init int init_syncookies(void) | |
31 | { | |
32 | get_random_bytes(syncookie_secret, sizeof(syncookie_secret)); | |
33 | return 0; | |
34 | } | |
c6aefafb | 35 | __initcall(init_syncookies); |
1da177e4 LT |
36 | |
37 | #define COOKIEBITS 24 /* Upper bits store count */ | |
38 | #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1) | |
39 | ||
245b2e70 TH |
40 | static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS], |
41 | ipv4_cookie_scratch); | |
11baab7a | 42 | |
714e85be | 43 | static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport, |
1da177e4 LT |
44 | u32 count, int c) |
45 | { | |
245b2e70 | 46 | __u32 *tmp = __get_cpu_var(ipv4_cookie_scratch); |
1da177e4 | 47 | |
2051f11f | 48 | memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c])); |
714e85be AV |
49 | tmp[0] = (__force u32)saddr; |
50 | tmp[1] = (__force u32)daddr; | |
51 | tmp[2] = ((__force u32)sport << 16) + (__force u32)dport; | |
1da177e4 LT |
52 | tmp[3] = count; |
53 | sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5); | |
54 | ||
55 | return tmp[17]; | |
56 | } | |
57 | ||
4dfc2817 FW |
58 | |
59 | /* | |
60 | * when syncookies are in effect and tcp timestamps are enabled we encode | |
61 | * tcp options in the lowest 9 bits of the timestamp value that will be | |
62 | * sent in the syn-ack. | |
63 | * Since subsequent timestamps use the normal tcp_time_stamp value, we | |
64 | * must make sure that the resulting initial timestamp is <= tcp_time_stamp. | |
65 | */ | |
66 | __u32 cookie_init_timestamp(struct request_sock *req) | |
67 | { | |
68 | struct inet_request_sock *ireq; | |
69 | u32 ts, ts_now = tcp_time_stamp; | |
70 | u32 options = 0; | |
71 | ||
72 | ireq = inet_rsk(req); | |
73 | if (ireq->wscale_ok) { | |
74 | options = ireq->snd_wscale; | |
75 | options |= ireq->rcv_wscale << 4; | |
76 | } | |
77 | options |= ireq->sack_ok << 8; | |
78 | ||
79 | ts = ts_now & ~TSMASK; | |
80 | ts |= options; | |
81 | if (ts > ts_now) { | |
82 | ts >>= TSBITS; | |
83 | ts--; | |
84 | ts <<= TSBITS; | |
85 | ts |= options; | |
86 | } | |
87 | return ts; | |
88 | } | |
89 | ||
90 | ||
714e85be AV |
91 | static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport, |
92 | __be16 dport, __u32 sseq, __u32 count, | |
1da177e4 LT |
93 | __u32 data) |
94 | { | |
95 | /* | |
96 | * Compute the secure sequence number. | |
97 | * The output should be: | |
e905a9ed | 98 | * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24) |
1da177e4 LT |
99 | * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24). |
100 | * Where sseq is their sequence number and count increases every | |
101 | * minute by 1. | |
102 | * As an extra hack, we add a small "data" value that encodes the | |
103 | * MSS into the second hash value. | |
104 | */ | |
105 | ||
106 | return (cookie_hash(saddr, daddr, sport, dport, 0, 0) + | |
107 | sseq + (count << COOKIEBITS) + | |
108 | ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data) | |
109 | & COOKIEMASK)); | |
110 | } | |
111 | ||
112 | /* | |
113 | * This retrieves the small "data" value from the syncookie. | |
114 | * If the syncookie is bad, the data returned will be out of | |
115 | * range. This must be checked by the caller. | |
116 | * | |
117 | * The count value used to generate the cookie must be within | |
118 | * "maxdiff" if the current (passed-in) "count". The return value | |
119 | * is (__u32)-1 if this test fails. | |
120 | */ | |
714e85be AV |
121 | static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr, |
122 | __be16 sport, __be16 dport, __u32 sseq, | |
1da177e4 LT |
123 | __u32 count, __u32 maxdiff) |
124 | { | |
125 | __u32 diff; | |
126 | ||
127 | /* Strip away the layers from the cookie */ | |
128 | cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq; | |
129 | ||
130 | /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */ | |
131 | diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS); | |
132 | if (diff >= maxdiff) | |
133 | return (__u32)-1; | |
134 | ||
135 | return (cookie - | |
136 | cookie_hash(saddr, daddr, sport, dport, count - diff, 1)) | |
137 | & COOKIEMASK; /* Leaving the data behind */ | |
138 | } | |
139 | ||
e905a9ed | 140 | /* |
5918e2fb FW |
141 | * MSS Values are taken from the 2009 paper |
142 | * 'Measuring TCP Maximum Segment Size' by S. Alcock and R. Nelson: | |
143 | * - values 1440 to 1460 accounted for 80% of observed mss values | |
144 | * - values outside the 536-1460 range are rare (<0.2%). | |
145 | * | |
146 | * Table must be sorted. | |
1da177e4 LT |
147 | */ |
148 | static __u16 const msstab[] = { | |
5918e2fb FW |
149 | 64, |
150 | 512, | |
151 | 536, | |
152 | 1024, | |
153 | 1440, | |
154 | 1460, | |
155 | 4312, | |
156 | 8960, | |
1da177e4 | 157 | }; |
1da177e4 LT |
158 | |
159 | /* | |
160 | * Generate a syncookie. mssp points to the mss, which is returned | |
161 | * rounded down to the value encoded in the cookie. | |
162 | */ | |
163 | __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp) | |
164 | { | |
aa8223c7 ACM |
165 | const struct iphdr *iph = ip_hdr(skb); |
166 | const struct tcphdr *th = tcp_hdr(skb); | |
1da177e4 LT |
167 | int mssind; |
168 | const __u16 mss = *mssp; | |
169 | ||
a0f82f64 | 170 | tcp_synq_overflow(sk); |
1da177e4 | 171 | |
5918e2fb FW |
172 | for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--) |
173 | if (mss >= msstab[mssind]) | |
174 | break; | |
175 | *mssp = msstab[mssind]; | |
1da177e4 | 176 | |
de0744af | 177 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); |
1da177e4 | 178 | |
aa8223c7 ACM |
179 | return secure_tcp_syn_cookie(iph->saddr, iph->daddr, |
180 | th->source, th->dest, ntohl(th->seq), | |
1da177e4 LT |
181 | jiffies / (HZ * 60), mssind); |
182 | } | |
183 | ||
e905a9ed | 184 | /* |
1da177e4 LT |
185 | * This (misnamed) value is the age of syncookie which is permitted. |
186 | * Its ideal value should be dependent on TCP_TIMEOUT_INIT and | |
187 | * sysctl_tcp_retries1. It's a rather complicated formula (exponential | |
188 | * backoff) to compute at runtime so it's currently hardcoded here. | |
189 | */ | |
190 | #define COUNTER_TRIES 4 | |
e905a9ed YH |
191 | /* |
192 | * Check if a ack sequence number is a valid syncookie. | |
1da177e4 LT |
193 | * Return the decoded mss if it is, or 0 if not. |
194 | */ | |
195 | static inline int cookie_check(struct sk_buff *skb, __u32 cookie) | |
196 | { | |
aa8223c7 ACM |
197 | const struct iphdr *iph = ip_hdr(skb); |
198 | const struct tcphdr *th = tcp_hdr(skb); | |
199 | __u32 seq = ntohl(th->seq) - 1; | |
200 | __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr, | |
201 | th->source, th->dest, seq, | |
202 | jiffies / (HZ * 60), | |
203 | COUNTER_TRIES); | |
1da177e4 | 204 | |
5918e2fb | 205 | return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0; |
1da177e4 LT |
206 | } |
207 | ||
1da177e4 | 208 | static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb, |
60236fdd | 209 | struct request_sock *req, |
1da177e4 LT |
210 | struct dst_entry *dst) |
211 | { | |
8292a17a | 212 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
213 | struct sock *child; |
214 | ||
8292a17a | 215 | child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst); |
1da177e4 | 216 | if (child) |
463c84b9 | 217 | inet_csk_reqsk_queue_add(sk, req, child); |
1da177e4 | 218 | else |
60236fdd | 219 | reqsk_free(req); |
1da177e4 LT |
220 | |
221 | return child; | |
222 | } | |
223 | ||
4dfc2817 FW |
224 | |
225 | /* | |
226 | * when syncookies are in effect and tcp timestamps are enabled we stored | |
227 | * additional tcp options in the timestamp. | |
228 | * This extracts these options from the timestamp echo. | |
229 | * | |
230 | * The lowest 4 bits are for snd_wscale | |
231 | * The next 4 lsb are for rcv_wscale | |
232 | * The next lsb is for sack_ok | |
8c763681 FW |
233 | * |
234 | * return false if we decode an option that should not be. | |
4dfc2817 | 235 | */ |
8c763681 | 236 | bool cookie_check_timestamp(struct tcp_options_received *tcp_opt) |
4dfc2817 FW |
237 | { |
238 | /* echoed timestamp, 9 lowest bits contain options */ | |
239 | u32 options = tcp_opt->rcv_tsecr & TSMASK; | |
240 | ||
8c763681 FW |
241 | if (!tcp_opt->saw_tstamp) { |
242 | tcp_clear_options(tcp_opt); | |
243 | return true; | |
244 | } | |
245 | ||
246 | if (!sysctl_tcp_timestamps) | |
247 | return false; | |
248 | ||
4dfc2817 FW |
249 | tcp_opt->snd_wscale = options & 0xf; |
250 | options >>= 4; | |
251 | tcp_opt->rcv_wscale = options & 0xf; | |
252 | ||
253 | tcp_opt->sack_ok = (options >> 4) & 0x1; | |
254 | ||
8c763681 FW |
255 | if (tcp_opt->sack_ok && !sysctl_tcp_sack) |
256 | return false; | |
4dfc2817 | 257 | |
8c763681 | 258 | if (tcp_opt->snd_wscale || tcp_opt->rcv_wscale) { |
4dfc2817 | 259 | tcp_opt->wscale_ok = 1; |
8c763681 FW |
260 | return sysctl_tcp_window_scaling != 0; |
261 | } | |
262 | return true; | |
4dfc2817 FW |
263 | } |
264 | EXPORT_SYMBOL(cookie_check_timestamp); | |
265 | ||
1da177e4 LT |
266 | struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, |
267 | struct ip_options *opt) | |
268 | { | |
4957faad WAS |
269 | struct tcp_options_received tcp_opt; |
270 | u8 *hash_location; | |
2e6599cb ACM |
271 | struct inet_request_sock *ireq; |
272 | struct tcp_request_sock *treq; | |
1da177e4 | 273 | struct tcp_sock *tp = tcp_sk(sk); |
aa8223c7 ACM |
274 | const struct tcphdr *th = tcp_hdr(skb); |
275 | __u32 cookie = ntohl(th->ack_seq) - 1; | |
1da177e4 | 276 | struct sock *ret = sk; |
e905a9ed YH |
277 | struct request_sock *req; |
278 | int mss; | |
279 | struct rtable *rt; | |
1da177e4 LT |
280 | __u8 rcv_wscale; |
281 | ||
af9b4738 | 282 | if (!sysctl_tcp_syncookies || !th->ack || th->rst) |
1da177e4 LT |
283 | goto out; |
284 | ||
a0f82f64 | 285 | if (tcp_synq_no_recent_overflow(sk) || |
1da177e4 | 286 | (mss = cookie_check(skb, cookie)) == 0) { |
de0744af | 287 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED); |
1da177e4 LT |
288 | goto out; |
289 | } | |
290 | ||
de0744af | 291 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV); |
1da177e4 | 292 | |
bb5b7c11 DM |
293 | /* check for timestamp cookie support */ |
294 | memset(&tcp_opt, 0, sizeof(tcp_opt)); | |
295 | tcp_parse_options(skb, &tcp_opt, &hash_location, 0); | |
296 | ||
8c763681 FW |
297 | if (!cookie_check_timestamp(&tcp_opt)) |
298 | goto out; | |
bb5b7c11 | 299 | |
1da177e4 | 300 | ret = NULL; |
ce4a7d0d | 301 | req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */ |
1da177e4 LT |
302 | if (!req) |
303 | goto out; | |
304 | ||
2e6599cb ACM |
305 | ireq = inet_rsk(req); |
306 | treq = tcp_rsk(req); | |
aa8223c7 | 307 | treq->rcv_isn = ntohl(th->seq) - 1; |
e905a9ed | 308 | treq->snt_isn = cookie; |
1da177e4 | 309 | req->mss = mss; |
a3116ac5 | 310 | ireq->loc_port = th->dest; |
aa8223c7 | 311 | ireq->rmt_port = th->source; |
eddc9ec5 ACM |
312 | ireq->loc_addr = ip_hdr(skb)->daddr; |
313 | ireq->rmt_addr = ip_hdr(skb)->saddr; | |
16df845f | 314 | ireq->ecn_ok = 0; |
bb5b7c11 DM |
315 | ireq->snd_wscale = tcp_opt.snd_wscale; |
316 | ireq->rcv_wscale = tcp_opt.rcv_wscale; | |
317 | ireq->sack_ok = tcp_opt.sack_ok; | |
318 | ireq->wscale_ok = tcp_opt.wscale_ok; | |
319 | ireq->tstamp_ok = tcp_opt.saw_tstamp; | |
320 | req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0; | |
1da177e4 LT |
321 | |
322 | /* We throwed the options of the initial SYN away, so we hope | |
323 | * the ACK carries the same options again (see RFC1122 4.2.3.8) | |
324 | */ | |
325 | if (opt && opt->optlen) { | |
326 | int opt_size = sizeof(struct ip_options) + opt->optlen; | |
327 | ||
2e6599cb ACM |
328 | ireq->opt = kmalloc(opt_size, GFP_ATOMIC); |
329 | if (ireq->opt != NULL && ip_options_echo(ireq->opt, skb)) { | |
330 | kfree(ireq->opt); | |
331 | ireq->opt = NULL; | |
1da177e4 LT |
332 | } |
333 | } | |
334 | ||
284904aa PM |
335 | if (security_inet_conn_request(sk, skb, req)) { |
336 | reqsk_free(req); | |
337 | goto out; | |
338 | } | |
339 | ||
e905a9ed YH |
340 | req->expires = 0UL; |
341 | req->retrans = 0; | |
342 | ||
1da177e4 LT |
343 | /* |
344 | * We need to lookup the route here to get at the correct | |
345 | * window size. We should better make sure that the window size | |
346 | * hasn't changed since we received the original syn, but I see | |
e905a9ed | 347 | * no easy way to do this. |
1da177e4 LT |
348 | */ |
349 | { | |
ffce9082 AE |
350 | struct flowi fl = { .mark = sk->sk_mark, |
351 | .nl_u = { .ip4_u = | |
1da177e4 LT |
352 | { .daddr = ((opt && opt->srr) ? |
353 | opt->faddr : | |
2e6599cb ACM |
354 | ireq->rmt_addr), |
355 | .saddr = ireq->loc_addr, | |
1da177e4 LT |
356 | .tos = RT_CONN_FLAGS(sk) } }, |
357 | .proto = IPPROTO_TCP, | |
86b08d86 | 358 | .flags = inet_sk_flowi_flags(sk), |
1da177e4 | 359 | .uli_u = { .ports = |
aa8223c7 ACM |
360 | { .sport = th->dest, |
361 | .dport = th->source } } }; | |
4237c75c | 362 | security_req_classify_flow(req, &fl); |
c4464921 | 363 | if (ip_route_output_key(sock_net(sk), &rt, &fl)) { |
60236fdd | 364 | reqsk_free(req); |
e905a9ed | 365 | goto out; |
1da177e4 LT |
366 | } |
367 | } | |
368 | ||
369 | /* Try to redo what tcp_v4_send_synack did. */ | |
d8d1f30b | 370 | req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW); |
4dfc2817 | 371 | |
1da177e4 | 372 | tcp_select_initial_window(tcp_full_space(sk), req->mss, |
e905a9ed | 373 | &req->rcv_wnd, &req->window_clamp, |
31d12926 | 374 | ireq->wscale_ok, &rcv_wscale, |
d8d1f30b | 375 | dst_metric(&rt->dst, RTAX_INITRWND)); |
4dfc2817 | 376 | |
e905a9ed | 377 | ireq->rcv_wscale = rcv_wscale; |
1da177e4 | 378 | |
d8d1f30b | 379 | ret = get_cookie_sock(sk, skb, req, &rt->dst); |
1da177e4 LT |
380 | out: return ret; |
381 | } |